CODA Requirement Intelligence Engine

CODA Requirement Intelligence

Turning operational signals into governed software execution.

CODA Requirement Intelligence turns operational signals into governed software execution.

From logs, grievances, policies, documents, requirements, and codebase evidence into classified control gaps, correlated candidates, implementation reality checks, and human-confirmed engineering decisions.

1.7M+Logs observed

177Requirements indexed

5Source reservoirs

25Pilot candidates

0Automatic mutations

YesCorrelation enabled

Current environment snapshot — includes live requirement and candidate batch counts from this environment. · File-backed candidate batches — not production authorization. · Safe fallback metrics; no full log scan on page load.

CODA breakthrough: one operating intelligence layer

Raw operational issue → system-control recommendation — not grievance tone rewriting.

Sources→ SourceSignal→ Classification + Risk→ Cross-source Correlation→ Control Gap Diagnosis→ Requirement Match→ Codebase Reality Check→ Guided Human Confirmation→ Evidence / Audit / Reconciliation

Example: “Employee may have approved payment improperly” → independent approval workflow, separation of duties, audit logging, exception review — not “employees should not steal money.”

Proof of scale

Current environment snapshot — safe fallbacks when live counts are unavailable.

1.7M+Logs observed

177Requirements indexed

5Source reservoirs

25Pilot candidates

0Automatic mutations

YesCorrelation enabled

Current environment snapshot — includes live requirement and candidate batch counts from this environment.
File-backed candidate batches — not production authorization.
Safe fallback metrics; no full log scan on page load.

Source reservoirs

New sources become adapters, not new intelligence engines.

Logs Live

Functional and operational patterns at scale.

Grievances Live

Control failures → system-control recommendations.

Policies Live

Obligations and approval language.

Documents Live

Procedure and requirement-like content.

Requirements Live

Indexed backlog — intent, not proof of implementation.

Google Drive / Box Future

Future adapters only — not a new intelligence engine.

SourceSignal architecture

Source

Origin of the signal — log, grievance, policy, document, requirement.

Classification

Interpretation — control gap type, risk domain, sensitivity.

Monitoring

Pattern tracking across time and routes.

Correlation

Relationships across sources — one cluster, one control story.

Decision

Confidence-scored recommendation — not approval.

Human review

Confirmation of high-impact decisions — not manual re-classification.

Multi-source correlation

Related signals across reservoirs converge on one control story — not duplicate tickets.

Policy Reimbursement requires approval

Grievance Payment control concern

Log Approval route fails repeatedly

Requirement Weak acceptance criteria

Document Finance procedure — manual approval

Codebase Partial workflow exists

Financial approval control gap

One cluster · one diagnosis surface

Control gap diagnosis

Automated diagnosis complete — humans confirm high-impact decisions.

87% Guided confirmation Financial-control confirmation required No strong requirement match Partial codebase implementation

Problem categoryPayment control concern

Likely control gapFinancial approval control gap

Confidence band87% — Guided confirmation

Criteria matched

Payment/control pattern detected
Grievance source with financial-control relevance
No strong requirement match found
Partial finance workflow in codebase
Missing separation-of-duties and audit-trail evidence

Suggested controls

Independent approval workflow
Maker-checker / separation of duties
Approval history log
Exception monitoring
Policy acknowledgment / reminders

Requirement matchNo strong match — closest weak matches listed for staff.

Codebase realityPartial implementation — approval view found; SoD tests missing.

Missing controls

Separation of duties enforcement
Audit trail tests
Exception monitoring

Human confirmation

System recommendation needs confirmation because this is a financial-control signal.

Codebase reality check

Requirements describe intent. The codebase reveals reality.

Code evidence is not human acceptance. The requirement row is not final proof of implementation.

Requirement intent

Independent payment approval should exist with separation of duties.

Codebase evidence found

finance payment model
approval view
reimbursement route

Missing evidence

Separation-of-duties enforcement
Audit trail test
Exception monitoring

Intent layer

What the requirement says should exist.

Implementation evidence

Models, views, services, templates, tests in repo.

Reality status

Partial / likely implemented / missing controls / needs tests.

Missing controls

Suggested controls not evidenced in code.

Next action

Improve requirement, tighten, attach evidence, or audit.

Confidence and decision thresholds

Score	Band	Meaning
90–100	Auto-ready recommendation	Still not approval. Low/medium risk only.
80–89	Guided confirmation	Strong diagnosis; sensitive domains still need explicit confirmation.
60–79	Human review required	Ambiguous requirement or codebase evidence.
Below 60	Defer / insufficient evidence	Gather more context before backlog action.

Financial, HR/grievance, security, and compliance signals always require confirmation before persistence or draft creation — even at high confidence.

Governance and safety posture

Evidence-only by default

RequirementPrompt artifacts — no automatic field mutation.

No automatic approval

GQI, Prompt Packs, and Control Cycle do not authorize production.

Production pilot guarded

Feature flags and staff-only routes — not autonomous rollout.

Sensitive source redaction

Pilot settings can redact HR/grievance text in reports.

Audit trail preserved

Evidence history and reconciliation decisions are retained.

governance_state_change: forbidden production_activation: blocked approval_gate: NOT_IMPLEMENTED

Core platform capabilities

Source Observatory

SourceSignal Layer

Multi-source Classification

Cross-source Correlation

Control Gap Diagnosis

Codebase Reality Check

Grouped Candidate Review

Evidence History

Implementation Audit

Reconciliation

GQI / Prompt Pack

Control Cycle

Technical architecture

Layer 1

Source adapters

Logs, grievances, policies, documents, requirements, future Drive/Box.

Layer 2

Universal SourceSignal taxonomy

Normalized identity, content, context, and source metadata.

Layer 3

Classification + risk

Content/context-based classification — not per-source duplicated engines.

Layer 4

Correlation clusters

Related signals converge into one control story.

Layer 5

Control gap diagnosis

Suggested system controls and confidence band.

Layer 6

Codebase reality checker

Requirement intent compared to models, views, services, templates, routes, tests.

Layer 7

Candidate review UI

Grouped staff confirmation — attach, defer, dismiss.

Layer 8

Evidence store / reconciliation

RequirementPrompt evidence and decision history.

Live evidence gallery

Primary evidence is the working route — screenshots are optional for PDF/offline export.

Requirements Intelligence Hub

Command center for backlog health, intelligence links, and AI_SYSTEM tooling.

Open requirements hub /requirements/

What to look for

Hub cards linking to candidate review, observatory outputs, and staff tools.

Executive: One front door for requirement intelligence — not scattered admin pages.

Governance: Review and evidence actions do not approve requirements or authorize production.

Technical: Read-only hub; no automatic requirement field mutation.

Candidate Review Queue

File-backed candidate batches from Source Observatory and correlation.

Open candidate review /ai_services/source-candidates/

What to look for

Batch list, grouped view toggle, unresolved vs attached counts.

Executive: Staff see patterns grouped by topic — not hundreds of duplicate rows.

Governance: Review and evidence actions do not approve requirements or authorize production.

Technical: Batches stored under var/ai_system/source_candidates; review_state only.

Current Pilot Batch

Active staff pilot batch when available; otherwise start from the queue.

Open candidate review queue /ai_services/source-candidates/

What to look for

Grouped review cards, human-readable actions, pilot guardrails banner.

Executive: Controlled pilot scope — not autonomous production rollout.

Governance: Pilot feature flags; production_activation remains blocked.

Technical: Dynamic batch link when review store has batches.

Grouped Review & Control Diagnosis

Automated control gap diagnosis on group and candidate detail.

Open grouped review (select a batch) /ai_services/source-candidates/?view=groups

What to look for

Automated Control Diagnosis block, suggested system controls, live route to codebase reality.

Executive: System completes diagnosis; humans confirm — not manual classification.

Governance: Diagnosis is not proof; recommendation is not approval.

Technical: control_gap_diagnosis service; optional RequirementPrompt evidence on attach.

Implementation Audit & Reconciliation

Compare requirement intent to codebase evidence; staff reconciliation queue.

Open reconciliation queue /ai_services/requirements/implementation-audit/reconciliation/

What to look for

Audit status, matched paths, missing controls, reconciliation decisions.

Executive: Closes the loop between backlog claims and code reality.

Governance: Reconciliation records decisions — does not auto-change spec status.

Technical: implementation_audit scanner + optional RequirementPrompt persistence.

Requirement Detail & Evidence

Open a requirement to view evidence history, GQI, and prompt packs.

Sample requirement CODA0008296 /requirements/8296/

What to look for

Evidence history list, AI_SYSTEM artifacts, unchanged requirement fields after evidence attach.

Executive: Evidence-backed requirements — not documentation theater.

Governance: GQI and Prompt Packs score only; they do not approve governance.

Technical: RequirementPrompt evidence types include control gap diagnosis when attached.

Staff Presentation Center

AI-generated pitch center for staff (separate from this flagship page).

Open presentation center /portfolio/requirements/

What to look for

Pitch generation form — login required; does not mutate requirements.

Executive: Optional staff tool for generated narratives.

Governance: Review and evidence actions do not approve requirements or authorize production.

Technical: Preserves /portfolio/requirements/ from PR 19A.

This Flagship Presentation

Curated demo command center with live links (you are here).

Demo mode deck /portfolio/requirement-intelligence-platform/demo/

What to look for

Live evidence cards, presenter notes, demo path — links over screenshots.

Executive: Proves the system is alive without waiting for PNG captures.

Governance: Presentation does not run pipelines or authorize production.

Technical: Offline-safe curated content + optional live metrics.

These working routes become the controlled staff pilot path: every step produces evidence, but no step silently approves, mutates, or activates production behavior.

Live demo path

Open working pages in order — each link is read-only or staff-governed review.

Requirements hub

Active requirements list and detail entry points.

/requirements/ Open

Requirement intelligence

Hub linking observatory, candidates, and staff tools.

/requirements/intelligence/ Open

Candidate review

Grouped candidate review and control gap diagnosis.

/ai_services/source-candidates/ Open

Reconciliation queue

Implementation audit findings vs requirement truth.

/ai_services/requirements/implementation-audit/reconciliation/ Open

Staff pitch center

AI-generated pitches (login) — separate from this flagship page.

/portfolio/requirements/ Open

This presentation

Flagship demo command center with live evidence links.

/portfolio/requirement-intelligence-platform/demo/ Open

CODA closes the loop

CODA closes the loop.

What happened in operations.

What policy required.

What the backlog claimed.

What the code actually implemented.

What humans approved.

This is the difference between documentation and governed execution.

CODA Analytics Technical